Connects decision-makers and solutions creators to what's next in quantum computing

Quantum generator outperformed classical GAN in computing the drug characteristics of generated compounds

John Potter

May 24, 2023

2 Min Read
A researcher looks over a set of test tunes
Quantum computing can create a training dataset for AI drug discovery. Getty

Insilico Medicine, a company focused on clinical-stage generative AI-driven drug discovery, has combined quantum computing and generative AI to explore lead candidates for drug discovery. The company intends to demonstrate the potential benefits of using quantum generative adversarial networks (GAN) in generative chemistry.

GANs are deep-learning neural networks that can produce highly effective generative models for drug discovery and design. They can be used to generate data designed to replicate the data distribution for a particular task. 

The standard GAN model consists of a generator and a discriminator. The generator uses random data noise as input and tries to mimic the data distribution over time, whereas the discriminator tries to distinguish between the fake and real samples. A GAN continuously trains until the discriminator cannot distinguish between the generated and real data.

In this study, researchers gradually replaced each component of MolGAN, an implicit GAN for small molecular graphs, with a variational quantum circuit (VQC). Researchers then investigated the quantum advantage in small molecule drug discovery and compared its performance with the classical computer counterpart.

The study demonstrated that the trained quantum GANs could create molecules that resembled those in a training set by employing the VQC as the noise generator. It also showed that the quantum generator outperformed the classical GAN in computing the drug characteristics of generated compounds.

Related:How Quantum Could Deliver Faster, Safer, Better Drugs: Q2B 2023 Paris

The study also found that the quantum discriminator of GAN, with only tens of learning parameters, could generate valid molecules. It could also outperform its classical counterpart, with tens of thousands of adjustable parameters, in generating molecule properties and the KL-divergence score – a metric that assesses the dissimilarity between the probability distributions of the generated molecules and the target distribution.

In connection with this endeavor, Insilico Medicine published a study in the Journal of Chemical Information and Modeling on accelerating drug discovery and development using breakthrough methods alongside new technologies such as generative AI and quantum computing.

About the Author(s)

Sign Up for the Newsletter
The most up-to-date news and insights into the latest emerging technologies ... delivered right to your inbox!

You May Also Like